Entity Candidate Extractor

Entities are the specific things a text refers to — people, organisations, products, places, standards — as opposed to the general words around them. Search engines have worked this way for over a decade: a page is understood less as a bag of keywords than as a set of entities and the relationships between them. Knowing which entities your page actually mentions is therefore a reasonable proxy for what a machine will conclude it is about.

This extractor is heuristic. It looks for capitalised sequences outside sentence-initial position, along with a few common patterns, which means it reliably catches proper nouns and reliably produces false positives — a capitalised word starting a sentence, a title-cased heading, or an emphasised term will all appear. Expect to discard a meaningful share of the output; the list is a prompt for review rather than a result.

Two things to do with the candidates. First, check that the entities central to your topic are actually named rather than gestured at — pages often discuss a concept at length while never stating the term someone would search for. Second, treat frequently repeated entities as internal linking targets and as candidates for their own page, since a concept substantial enough to recur across several articles usually deserves a definitive one.